Illuminate: Depression diagnosis, explanation and proactive therapy using prompt engineering

Traditional methods of depression detection on social media forums can classify whether a user is depressed, but they often lack the capacity for human-like explanations and interactions. This paper proposes a next-generation paradigm for depression detection and treatment strategies. It employs three Large Language Models (LLMs) - Generative Pre-trained Transformer 4, Llama, and Gemini, each fine-tuned using specially engineered prompts to effectively diagnose, explain, and suggest therapeutic interventions for depression. These prompts are designed to guide the models in analyzing textual data from clinical interviews and online forums, ensuring nuanced and context-aware responses. The study utilizes a few-shot prompting methodology for the Diagnosis and Explanation component. This technique is optimized to provide DSM-5 based analysis and explanation, enhancing the model’s ability to identify and articulate depressive symptoms accurately. Models engage in empathetic dialogue management, guided by resources from Psychology Database and a Cognitive Behavioral Therapy guide, and fine-tuned using Chain of Thought and Tree of Thought prompting techniques. This facilitates meaningful interactions with individuals facing major depressive disorders, fostering a supportive and understanding environment. The research innovates in case conceptualization, treatment planning and therapeutic interventions by creating the Illuminate Database to guide the models in offering personalized therapy. The quantitative analysis of the study is demonstrated through metrics such as F1 scores, Precision, Recall, Cosine similarity, and ROUGE score across different test sets. This comprehensive approach offered through a mobile application prototype, with established psychological methodologies showcases the potential of LLMs in revolutionizing diagnosis and treatment strategies.